Birth Nativity, Citizenship, and Gender Difference of Immigrant. Scientists/Engineers Earnings

Birth Nativity, Citizenship, and Gender Difference of Immigrant Scientists/Engineers Earnings Yuying Tong University of North Carolina at Chapel Hill ABSTRACT In this paper, I employ a random-effect growth curve model on a longitudinal data set of scientists/engineers to model the earning differences by birth nativity, citizenship and gender. Four waves of SESTAT data were arranged into a pooled-cross section time series so that repeated measures of scientists/engineers for each individual in a two-year interval could be used for analysis. The results show that an unobserved random effect explained nearly 30 percent of the variance on the overall earning differences across individuals. We also find that overall foreign-born scientists are not necessarily at a disadvantage for both overall earning and earning growth rate. Citizenship status, however, plays a significant role on foreign-born scientists/engineers earning disadvantage, but not in the earning growth rate. Women do experience disadvantages in both overall earning and on earning growth rate, but the evidences of foreign-born woman scientists are more disadvantaged than their native counterparts just exist in those foreign-born without citizenships. Introduction Due to the large proportion of immigrants and concerns on their adaptation to U.S., a sizable literature has examined the integration of immigrants into the U.S. society. How immigrants perform in the U.S. labor market has been one of the central questions in these studies (Borjas, 1994). Different answers for this question underlie much of current debate on cost and benefit for the host country. Using Census data, the earliest influential work by Chiswick (1978) indicates that relative earnings of immigrants grow fast and eventually overtake the earnings of native workers. Borjas (1985, 1989, 1994), however, suggests that non-random emigration and quality differences across immigrant cohorts would bias this across-section estimates. He finds that the assimilation rate measured in cross-section studies partly due to a decline in the quality of immigrants admitted to United States since 1965, after the Immigration and Naturalization Act eliminated national origins quotas. 1

It is well known that the relative skills of immigrant cohorts declined substantially when the national origin mix shifted away from the traditional European source countries toward Asian and Latin countries due to the 1965 Immigration Act. Those immigrants are also characterized by being more likely entering to reunite with kin than on the basis of their occupational skills (Duleep and Regets 1996) since one key factor is that an immigrant who becomes a U.S. citizen is allowed to sponsor family members for obtaining visas by this law change. As a consequence, the immigrants in the United States are fairly heterogeneous with respect to ethnicity, social class, and other characteristics correlated with economic stratification. Using census data, previous studies on immigrant earning has taken the immigrant population as a whole or only chosen the sub-sample of men to study the immigrant labor market outcome. Less attention has been paid to group differences across gender, social class etc. For example, Gender gap has been paid much less attention when studying immigrant earnings, and the earning difference between immigrants and their native counterparts for low skilled labor workers might be very different from the high educated groups. There is much work which can be done to study the gender differences of different immigrant groups. This paper will only focus on the high educated group of scientists/engineers to examine the gender difference on average earnings and earning growth using a longitudinal dataset of repeated measures of individuals. For many decades, highly skilled immigrants have pursued a higher premium on their education and skill by coming to the United States. Only recently well-educated immigrants have been brought to the attention of the general public and policy makers (North 1995), partially because they have high level of productivity in host country and 2

there are less public concerns on the costs of host country (Vernez 1997). The size increased for the well-educated group dramatically in 1990s; 13 percent of all college graduates in the U.S. civilian labor force was foreign-born in 2000, and over one-third arrived in the 1990s. However, the labor market for high educated population became tighter and tighter, and more economic concerns and apprehension arises among natives. How does this group perform in U.S. labor market compared with native born? Are there any group differences in the labor market performance of different race/ethnicity background, gender groups? Answers to these questions will help increase the knowledge about the costs and benefits of having this group in U.S. Previous research has already shed some lights on this subject matter (Bojars 1989; North 1995; Tang 1993; Goyette and Xie 1999; Xie and Shauman 2003). However, most of the studies, except Goyette and Xie (1999), focused on inequalities in labor market outcomes by nativity or generation but overlooked the role of gender. According to Pedraza (1991), the experience of immigration profoundly impacts the both public and private lives of women. Compared with men, women are more likely to accompany their husbands, or carry along children when they migrate. As Houstoun (1984:919) stressed, women generally migrate to create or reunite a family. Women s migration is more likely seen as the secondary movements generated by the original migration of economically motivated by males. Hence, different experience of immigration might have different impacts on men and women s labor market choice and outcome. As far as immigrant scientists/engineers are concerned, with more autonomy and higher self-esteem, one would expect that the pattern of gender difference among the 3

higher educated immigrant group might be similar to natives. Using 1990 census PUMS data, Goyette and Xie (1999) examined this hypothesis and found that foreign-born female scientists/engineers make about 4.7% less than the combination of other scientists/engineers after considering the impact of immigrant status. However, more input is still needed in this area, especially using longitudinal data. Women s roles in public life and private life change over time, for example, their family responsibilities change over their life course. Using longitudinal data would catch the information of life changes which could impact labor market outcomes in a person s life. Further more, some unobserved factors may correlate with the choice of immigration to U.S., naturalization process, which might affect their performance on the labor market. So a longitudinal data set and more advanced research method need to be used to address earning differences from average earning and earning change overtime for a person. In this study, I will use repeated measures of the same individuals over different survey years to examine the effect of immigrant status, which includes both birth nativity and citizenship, on the earnings of scientist/engineers by gender. To better solve the problem of unobserved factors, I employ random-effect growth curve models. Random effect growth curve models will give not only the effects from time invariants and timevarying covariates, but also unobserved random effect. Hence, I can determine not only if there is an effect of nativity and citizenship on the scientists/engineers earning and if there is difference between men and women, but how much the effect and difference is, if they exist. Literature Review 4

Prior to the 1965 Amendments to the Immigrants and Nationality Act, immigrants to the United States were regulated by numerical quotas based on the ethnic population of the United State in 1920. This encouraged immigrants from European countries and restricted immigrants from Asia and Latin America. After 1965, the Immigration Act allows more individuals from third world countries to enter the US (including Asians, who have traditionally been hindered from entering America); it also entails a separate quota for refugees. Skill/Professions or the relatives of U.S. Citizens are issued visas to come to U.S and countries of origin are no longer significant barriers for the immigrants. As a result, the new flow of immigrants originates mostly in Asia and Latin America, and they are more mixed on their race/ethnicity. Since one of fundamental shifts of the mechanism to accept immigrants whose skills are what U.S. labor market demands or who have kinships in the United States. As a result, internal heterogeneity might be the most significant characteristic in the current immigrant population in the United States. Migrant streams often alter the composition of places with respect to ethnicity, social class, and other characteristics correlated with economic stratification (Cobb- Clark, 1993). Ethnicity often defines the boundaries for social and cultural interaction. Previous studies have concluded that it is no longer necessary to provide ad hoc explanation of why the U.S. earnings of immigrants from different countries tend to exhibit so much variation. The economic theory of immigration suggests that this variance can be explained in terms of the economic and political conditions that guided the nonrandom sorting of persons across countries at the time of migration (Borjas, 1989). However, compare to this, earning differences among other different immigrant groups are far less clear. 5

The study on the immigration duration and earnings has led to lots of debate in the literature. The earliest work by Chiswick (1978) used the cross-section of census data and found that the earnings of foreign born persons immediately upon arrival are likely to be lower than the earning of comparable natives. Overtime, however, since immigrants have lower earnings, they also have higher incentives to invest human capital than natives. Immigrants earning can be expected to rise relatively fast as the returns to human capital investments are realized. The catch-up earnings profiles reflect the assimilation or adaptation of immigrants to the host country s labor market (Chiswick, 1978, Becker, 1975). This implies that immigrants will be self-selected not only on the basis of wage levels, but also on the basis of wage growth. However, the conclusion that immigrants have relatively high earnings growth has been challenged on both empirical and theoretical grounds (Duleep and Regets 1997). Borjas (1985) argues that cross-sectional framework used in Chiswick s study might bias the estimates because nonrandom emigration and immigrant cohort quality changes over time. He argues that if there has been a decline overtime in the earnings ability of immigrants, then the assimilation effect measured in cross-sectional studies could be spuriously inflated by declining immigrant earnings ability. In his other studies, Borjas found that immigrants initial wages adjusting for education and age, have decline overtime (Borjas 1992). Studies using cross-sectional census data can not sufficiently solve the problems of bias due to the cohort quality changes and emigration. In addition, an internal assumption in these studies is that immigrants and natives approximately have similar occupational composition in the United States. However, it is well known that the current 6

immigrants in the U.S. are more likely to be in both lower tail and upper tail of occupational prestige distribution than being in the middle. Since the opportunities sets faced by high educated immigrants are different from other group immigrants, it would be more valuable to compare a particular immigrant group with their native counterparts on the earning patterns than comparing the undifferentiated mass with all natives. An examination from both average earning and earning growth will fully capture the immigrant earning pattern. As an important part of scientific workforce in the United State, the immigrant scientists/engineers are relatively less heterogeneous with regard to their human capital within group. Although a relatively small proportion in the immigrant population, the number of foreign-born scientists/engineers keeps climbing with each passing year, and a higher percentage of the college-educated foreign born holds post-graduate degrees than the native born, with 43.6 % percent holding a master s, professional, and/or doctoral degrees, compared to 35.2 % of the native-born. Its important impact on U.S. talent labor market has led to heated discussion (Goyette and Xie, 1999), and overall, the research has drawn conclusion on immigrant scientists/engineers to either displacement (North, 1995) or discriminations (Tang, 1993, 2000). From the perspective of displacement, North (1995) argues that there are two groups of foreign-born scientists/engineers. One group is those who come in one at a time, study at U.S. graduate schools, secure advanced degree and then, in large number, stay in the U.S. The other group is those who enter already holding a degree and participate in U.S. labor market without U.S. education. Because there are so many of them obtain degrees from U.S. graduate school, immigrant scientists occupy positions 7

that might otherwise be taken by women and native-born minorities. In other words, this means that U.S. need not exert itself to expand the efforts to get Americans, particularly women and minorities to enter science and engineering graduate school. According to the displacement perspective, immigrant scientists/engineers take the downward pressure on the payment structure of science and engineer field. In contrast, the discrimination perspective (Tang, 1993, 2000) posits that immigrant scientist face unfair treatments in the U.S labor market. In studies of Asian scientists in U.S. labor market, Tang argues that there is ample evidence to show that Asians, regardless of gender, continue to have a lower level of income and career status than Caucasians with comparable training and qualifications (Barringer, Takeuchi, and Xenos 1990; Chu 1988; Hirschman and Wong 1984; Nee and Sanders 1985; U.S. Commission on Civil Rights 1988). One reason could be that those minority scientists/engineers are more likely to be confined to employment in the periphery of profession (where opportunities are scarce) and, in turn, suffer from significant income loss and downward occupational mobility (Wu 1980; Sung 1976; Villones 1989). Numerous studies have examined the adverse effect of nativity status on the earnings of Asian immigrants (Hirschman and Kraly 1988; Hirschman and Wong 1981, 1984; Nee and Sanders 1985; Poston and Jia 1989), and less research have paid attention to other underrepresented immigrant groups. Comparing these two perspectives, both suggest that immigrant scientists/engineers as a whole are in disadvantaged compared with native-born scientists/engineers in the labor market. However, in a further examination, one couldn t see they are different on what is the ultimate reason of disadvantages: it could be due to either their citizenship 8

status, for example, they are foreign-born but holding U.S citizenships, or simply due to their foreign-born status no matter if they have U.S. citizenship. Foreign-born who are holding U.S. citizenship is naturalized U.S. citizen. In general they have lived in U.S. for at least 5 years, and they have strong claim to rights in U.S. society. In theory they are entitled to the same privileges as native-born citizens (Massey and Bartley 2005). Thus, if it in the latter situation, foreign-born scientists/engineers would still be in the disadvantaged after controlling for citizenship. I have not seen any previous research that look into this point. In this paper, I test this by combining both birth nativity and citizenship status to indicate immigrant status to examine the earning pattern in my model. In previous studies, one common problem is that the role of gender is often omitted. According to Pedraza (1991), gender plays a central role in the decision to migrate and the composition of the migration flows, with the consequences that composition holds for the subsequent form of immigrant incorporation. There are more men than women among foreign-born, college-educated workers in the scientific workforce. In 2000, fifty-eight percent of foreign-born, college-educated workers are men, and this percentage was even higher among college-educated migrants who have arrived since 1990. In comparison, men constitute 53 percent of the native-born, collegeeducated workforce. However, women represented only 21.3 percent of the scientists and engineers being admitted with permanent resident status in 1993. Women, as Pedraza (1991) argued, are more likely to be secondary movements of males. So without consideration of gender differences in the study, the picture of immigrants/scientists would be incomplete. As one needs a long term investment and engagement to pursue 9

high degree and work in scientific field, women with special role related to family might make the impact of migration experience on labor market outcome different from the male peers. Thus, inattention to gender difference may result in an inaccurate characterization of the experiences of immigrant scientists and engineers. For example, Goyette and Xie (1999) argue that married women who work in scientific field may be more likely to come to U.S. as secondary immigrants of their husbands. Much more women than men migrate to get unification with their spouses, so their choice on career is greatly limited even if they might have similar education background on the time of migration compared to their male counterparts. Motivated by this, Goyette and Xie, using the 5 percent PUMS data from the 1990 U.S. census, first time systematically studied the effect of immigrant status on labor force participation, earnings and promotion of immigrant scientists/engineers. Their regression result show that foreign born female scientists/engineers earn about 5 percent less than all other group scientists/engineers. They argue that family responsibilities mainly account for women earning disadvantages. However, the measurement of earning in Goyette and Xie s study is the annual earnings in 1989, and the results from cross-sectional PUMS data are static and as such have no bearing on dynamic process such as the earnings change that could capture the a full picture of gender differences of earning in a person s life course. In addition, crosssectional data might provide biased estimates on the association of immigration status and mean earnings. The potential bias is due to the well-known problem that a single cross-section regression cannot differentiate the effects between migration experiences and individual characteristics. In this context, immigration status captures the difference in earnings among a typical immigrant scientist/engineer and a native scientist/engineer, 10

while the individual effect captures the productivity (or ability), ambition difference across different individuals. Since individual effects such as ability, ambition, intelligences etc might be correlated with immigration status, using immigrant status as predictor to study earnings outcomes in a cross-section data might bias parameter estimation. That is, in cross section analysis, if immigrant scientists/engineers have no earning disadvantage compare to natives; it may be due to their higher ability or higher career ambition which compensates for the shortfall as immigrants. To deal with the problem, in this paper, I use a repeated individual data set and a random-effect growth curve model to control for unobserved variables such as ability, ambition and intelligent factors. The exact amount of unobserved individual effect is determined by factor of personal characteristics, that is, the personal characteristics that would not change over time. In sum, this paper will use repeated measures of individual from longitudinal data to examine the effect of birth nativity and citizenship on both mean earnings and earning growth. Gender difference is one of important mission in this paper to fully capture the women immigrant scientists/engineers earning profiles relative to others. Methodology Data I take the longitudinal part of Scientists and Engineers Statistical Data System (SESTAT) integrated data as my analysis dataset. SESTAT is a database of the employment, education, and demographic characteristics of the nation s scientists and engineers. The SESTAT integrated the survey data from three component surveys, which includes the 11

National Survey of College Graduates (NSCG), the National Survey of Recent College Graduate (NSRCG) and the Survey of Doctorate Recipients (SDR). All of those surveys have been sponsored every two years since 1993 by the National Science Foundation. In this paper, I use the integrated databases for 1993, 1995, 1997 and 1999. Although the period from 1993-1999 is a relatively short period, the data collected earlier than 1993 is not comparable with data collected in 1990s due to the mechanisms of data collection changes. One also might notice that 1990s is a period for U.S. economic expansion, so I assume the period effect for immigrant scientists/engineers are same for immigrants vs. natives and men vs. women. The target population of SESTAT includes residents of the United States with at least a bachelor s degree and who, as of the survey period (April 15 for each survey year), was non-institutionalized, 75 years of age or less and either educated in a science or engineering (S&E) field or working in an S&E field. As a result, although some scientists/engineers dropped out from the follow-up surveys and some fresh graduation were recruited; the majority of the respondents still have been measured repeatedly over each survey year, with more than 50,000 individuals surveyed in all fours waves. The response rates vary across survey components and across survey years, range from 77 percent to 95 percent. Although this data set was integrated from different survey components and years, more than 90 percent measurements are exactly same across different surveys. It is valuable to arrange the longitudinal data into a pooled-cross section time-series in which the unit of analysis is an individual in a particular survey year. 12

Measures Dependent variable: The dependent variable for this study is the natural logarithm of annual salary. This variable was constructed from the salary of principle job the individual holds in the survey reference week (April 15 in the survey year) before deductions. Although it is possible for a scientist/engineer to take secondary job, the total earned income from all jobs was not surveyed in year 1993. Values are top-coded at 150,000 and rounded to the nearest thousand. In addition, non-zero values are bottom-coded, and values greater than zero but less than 5000 are assigned the value "4999". Since salary is a form of recognition for professional contributions and a measure of worth in the scientific community (Long, 2001), it usually accumulates over time. In this study, each person will have at least one and at most four measurement of earning. Independent variables: Immigrant status: As argued in the literature part, immigrant scientists/engineers may be in earning disadvantages because of both displacement theory which argues immigrants are more likely to accept low paid job and discrimination theory which explains immigrants being in labor market disadvantages because of discrimination. Immigrant status is the key predicator of this study, which is measured using both birth nativity and citizenship. Immigrants who own U.S. citizenship have a much broader economy opportunities than immigrants without citizenship (Yang 1994). For example, they enjoy lots of benefits on education and job choices. Previous studies always use birth nativity to indicate immigrant status. However, using foreign-born versus native-born as a dummy 13

variable to indicate immigrant status will lose some important information. To differentiate what is the ultimate reason of disadvantages, for example, foreign-born status (not well assimilated into U.S. society) or citizenship status (policy discrimination toward non-citizen), I will use a categorical variable of three groups as foreign-born without citizenship (FBWOC), foreign-born with citizenship (FBWC) and native-born (NB) to indicate the immigrant status. Since a non-citizen immigrant might be naturalized later on in his work life, so a citizenship will be a time-varying variable in the model. Demographic variables: Age could be a factor contributed to earning differences between foreign born and native born scientists/engineers. Age could also be a factor contributed to earning differences between males and females since women enter into the science and engineer is relatively new (Long, 2000). The age composition of immigrant scientists/engineers might be different from their native counterparts. Since almost onethird immigrants worked in natural and social science, engineering, and computer-related occupations arrived between 1990 and 2000, immigrant scientists/engineers are generally younger than natives. In the dataset, the earliest cohort was born before 1929 and the latest was born later than 1970. I use the birth cohorts to present the age differences, which makes the age group a time-invariant variable. I grouped them into five categories: 1965 or later, 1955-1964, 1945-1954, 1935-1944 and 1934 or earlier. Gender is measured with a dummy variable of male and female. Race/ethnicity is another variable that might be confounding with the earning differences between foreign-born and natives. Race is recoded into the broad categories including White, Asian and Other. Other includes Hispanic, African Americans, and other under represented minority. 14

Human capital variables: Although the sample in this study is relatively a homogeneous group with all of them holding high education degrees, the post-graduate education degrees holders are usually significantly different from those holding bachelor degrees. Compare with native-born, foreign-born college graduates are more likely to hold post-graduate degrees. Female college graduates are less likely to hold post-graduate degree compared with males. In this study, I recoded a series of dummy variables to indicate the education, which includes bachelor, master, professional, and other degrees as well as doctoral degrees. It is well known that work experiences is an important explaining variable in individual earning, but a direct measure of work experience in the data set is not available, so I use age as a proxy. At the same time, I use dummy variable (full-time versus. part-time) to indicate employment status, since women are more likely to attend part-time job. All the human capital variables might be changed over the life course, for example, a person might hold a bachelor degree at 1993, but hold a master degree in 1997. Work field: It is well known that the distributions of men and women within high educated group are extremely uneven (Jacobs 1996). In a cross-national study of high education group, Charles and Bradley (2002) concludes that female underrepresented in engineering, math/computer science (and to a lesser degree, natural science); female overrepresented in education, humanities, and health fields; and approximate gender parity in the social science. They argue that this is consistent with the culture-centered and human-capital accounts, since both of which predict it is characterized by functional or symbolic proximity to traditional female roles (Becker 1991; Reskin 1993). Because of segregation of fields by gender in high educated population, it is necessary control work 15

field to net out this confounding factor. I group the scientists/engineers into five wellestablished categories for scientists/engineer occupation: computer and mathematical, life and related science, physical and related science, social and related science as well as engineering. Employment sector also is an important factor that impacts earning. Usually industrial offers higher wage than academia or government (Peek, 1995, Goyette and Xie 1999). Employment sector includes four categories: industry, academia, government and other. Work field variable may also be changed for a particular over time. Family responsibility: Women in contemporary U.S. on average, get better grades in school, take math and science and science classes in the same rate, and earn roughly the same number of bachelor s degree in science and engineering as men. But in the career path, because of childbirth, cultural norms and social expectation, they tend to become scarcer in the highest ranks. Among them, childbirth might be the most significant factor which barrier women s careers. Although there is no evidence that immigrant women scientists/engineers are having more children, the effect of having young children for those women might be bigger than native-born women due to their less social and kinship network support. In this paper, a time-varying variable of number of young children under a certain age will be included for both male and female and to see if it only impacts women s earning. Statistical Approach Since individuals were repeatedly surveyed across years, and the motivation of this study is to examine both amount of earnings and slope of earning growth over the survey years, random-effect growth curve model is used. Random effect regression models with a 16

random intercept account for the individual differences in initial earning difference (1993 in this study) when the survey started, and random effect coefficient models account for the individual differences on the slope of earning growth. This paper will analyze the effects of gender, immigration status (including both nativity and citizenship) on earnings over time taking into account the change in employment changes, education changes, and citizenship status changes, changes of the field and employment sector as well as family responsibility. Most of previous studies on earning which use longitudinal data choose the fixed effect models (England et al, 1988, 1996) to control the unobserved effects. The fixed effects models based on longitudinal data allow us to control for unmeasured effects that are constant across repeated measures over time (Guo and Hipp, 2003). For example, England et al (1996), studied the effect of gender composition on starting wages in an organization, pooled across all job spells for each worker to control for such unmeasured and unchanging personal characteristics as intelligence, preferences resulting from early socialization, life cycle plans, and unmeasured human capital. Compared with fixed effect model, random effect could give an evaluation of variation of outcome based on both individual and measurement levels in a longitudinal data. In addition, random effect growth curve model can give the parameter estimates of time-invariant covariate effects, while in fixed effect model; those variables will be swept out from the model since fixed effect model controls the constant effect in the model. In this study, gender and birth cohorts are seen time invariant variable, and the parameter estimates for them will be important to know to answer the research questions in this paper. In a random effect 17

model, the effects of unobserved variables are estimated side by side with the effects of observed variables. Random-effect growth curve model can be represented through a two-level hierarchical model. At level 1, each person s earning is represented by an individual growth curve trajectory that depends on a unique set of parameters. These individual growth parameters become the outcome variables in a level-2 model, where they may depend on some person-level characteristics (Raudenbush and Bryk, 2002). In my analysis, each individual is measured at least one time and at most four times, and a random-effect growth curve model may be expressed as follows: Y p = β + β time + j j ij β kj x + ij 0 1 ijk k =1 e ij Level 1 (1) β = β Level 2 + 0 j 00 u0 j β = β Level 2 + 1 j 10 u1 j (2) (3) Y Y p = ij β + β time + j ij β kj x + ijk u + 00 1 0 j k = 1 p = ij β + β time + ij β ij x + ijk u + j u jtime + 00 10 0 1 ij k = 1 i =0, 1, 2, 3; j=1, 2, 3, 4, 5, 6 e ij e ij Combined (1) & (2) Combined (1), (2) & (3) (4) (5) Equation (4) is a random-intercept only model, and equation (5) includes both random intercept and random coefficient. In this model, y is the annualized salary for ij 18

individual j at timei, β 00 is the intercept (here it represents the annualized salary at time 0(that is in 1993 survey time). u j is the individual-specific random effect, and eij is the measure-specific random effect or the OLS-like error term. The standard assumptions are that u j and eij are mutually independent N (0, σ 2 u ) and N (0, σ 2 e ) random variables, where σ 2 u and σ 2 e are between-individual variance and within-individual variances, respectively. P is the number of covariates, and β kj is the coefficient of covariate k. In addition to what is captured by the observed variables (covariates), this model assumes that each measure is subject to two effects. One is unique in each measure (e ij ) and the other (u ) is same for all measures of an individual, but differs by individual. The j quantities of σ 2 u and σ 2 e are these two effects variances, respectively. A large-between individual variance indicates large differences in earnings across the individual. The addition of the individual-specific random effect u j is the only difference between this kind of model and typical OLS model. The assumption is that controlling for the individual-specific random effectu, the multiple measures of the individual will be j independent. Result Table 1 presents the means or percentages of variables by survey year and table 2 displays the means or percentages of variables by gender and by survey year. Gender, race and birth cohorts are time-invariant variables. The immigrant status, employment status, recent degree and field, work sector and number of children are all time-varying 19

variables across years. For example, it is possible for a person who is a Bachelor degree holder in 1993 to become a Master degree holder in 1997. By using time-varying covariates, one can better capture the covariates effect on the labor market outcome. According to table 1 and table 2, overall the annual salaries for all scientists/engineers are growing over time. There are about 20 percent foreign-born scientists/engineers in the sample. About 12 % scientists/engineers are Asians, which is a largest proportion of minority in the high educated labor market. In the sample, I exclude those people who were out of labor force, and more than 90 % in the rest of respondents are full-time labor force participators. Women are more likely to be part-time workers than men. Women are more likely to have bachelor degree than men, and men are more likely to hold doctorate degrees. The proportion of people in engineer among men is higher than the proportion of people in engineer among women, and reverse is the life and science field. The proportion of having young children under age 12 is higher for male than for female scientists. Table 3 shows the correlation matrix for the earning across the survey year. Since salary is a form of recognition for professional contributions and a measure of worth in the scientific community (Long, 2001), it usually accumulates over time. In other words, measures close to each other tend to be similar to each other. So there is a possibility of autoregressive structure over time (Guo and Hipp, 2003). This is specially the case for the earnings growth. In addition, due to the attrition and missing, measurements for some individuals in some particular years are not available, so data is not balanced; to solve this problem, a covariance structure that accounts for the first-order autoregressive process (Littell et al, 1996) is used to estimate the parameters. 20

The model (not shown) indicates that there is about 28% variances are explained by unobserved random effect (individual characteristics) after controlling for the observed covariates. In table 3, time has a positive effect on the earning, which indicates that the overall earning is growing over time. Women scientists/engineers are in disadvantages compared with men since women earn about 16% less than men. As I expected, citizenship plays an important role on immigrant scientists/engineers earnings. The results show that foreign-born without citizenship is really in disadvantages compared with native-born, but those foreign-born with U.S. citizenship are not. This case holds for both male and female. However, female foreign-born without citizenship earn much less comparing with other group females. They earn 13% less than native women, but males just learn 8% less than native men. Those findings suggest that the disadvantages are mainly resulted from the non-citizenship status, that is, the policy discrimination to the foreign-born people without citizenship in the labor market. In addition, women are particularly in disadvantages. Asian men earn less than their white counterparts, but this pattern does not exist for women. The birth cohorts for 1935-1944 and 1945-1954 are higher than birth cohort of 1934 or earlier, the less earning of older scientists/engineers might be due to their skill differences. This indicates that it is not necessarily the older, the more on earnings for scientists/engineers. It is not surprising that full-time jobs can earn much more than parttime jobs. Also it is expected that all post-graduate degrees can make people earn much more money than bachelor degrees, this holds for both men and women. Comparing with social sciences, all other field can make people make more money. The only exception is 21

the life science for women. As other argued before, people work in academia and government earn less than in industry. As I mentioned before, random intercept model captures the growth of earning overtime, for example, at time 0 (1993), the intercept represents the logged earning at time 0. For the logged earning at time 1(1995), the earning would be the intercept plus the parameter estimate multiplying by 1, and multiplying by 2 and 3 in time 2 and 3 for a typical person, respectively. Although this can reflect the overall earning differences, the earning growth rate cannot be represented in a random-intercept only model. Fortunately, a random coefficient model can be used to detect the growth rate or slope. Table 5 gives the results for this model. By comparing the parameter estimations for predicting intercept and predicting slope, I found that female have both intercept earning disadvantages and slope disadvantages, which means that women are in the disadvantages not only on the amount of earnings, but also in their earning growth pace. Foreign-born scientists/engineers without citizenships are in disadvantages on the amount of earning (intercept), but they have a faster growing pace relative to their native counterparts for males. Still, foreign-born with citizenship have no differences compared with natives. For the control variables, Asians are in disadvantages regard to intercept, but not in grow pace. Still, birth cohorts are nonlinear on intercept compared with the oldest cohorts, but the earning growth is faster when they are younger. Doctorate degree holders have advantages in both intercept and slope compared with Bachelor for women, but the slope change is negative for men. Although academia and governments jobs have disadvantages regard to the amount of earnings, they have faster growth rate compared with industry jobs. 22

According to the statistical results described above, women are indeed in earning disadvantage for both intercept and slopes, and foreign-born women without citizenship are particularly in disadvantages. What would be the degree of disadvantages for foreignborn women scientists/engineers relative to foreign-born men? Table 6 presents the results for random growth curve models of immigrant sample only. From the table 6, one can find that foreign-born women make 14.5% less than foreign-born men. And the growth rate is 2 % less than men. Foreign-born scientists/engineers without citizenship earn 11.3% less than those naturalized foreign-born, but they have 1.8% higher growth rate. Conclusion Are the immigrant scientists/engineers in the earning disadvantage? What is the difference between males and females? Are female scientists/engineers are particularly in disadvantages? In this article, I use the repeated measures of scientists/engineers to advance the previous study by conducting random-effect growth curve model. This model allows one to control the unobserved random effect such as ability, early socialization, intelligences, and personality for a particular individual. The results show that unobserved random effect explained near 30 percent variance on the earning differences. I found that not all foreign-born scientists/engineers are necessarily in earning disadvantages, but the citizenship status really plays a negative role on the amount of earnings. Foreign-born scientists/engineers without citizenship earn less than both natives and naturalized foreign-born. However, when the models allow the slope to change across the individuals, the earning growth rate of foreign-born scientists/engineers 23

without citizenship became positive, although they are still in earning disadvantages regard to the amount of earnings. About the gender differences, I found that women scientists/engineers are in disadvantage in both the amount of earning and growth rate. Foreign-born women are in disadvantages comparing with both native women and foreign-born men. Reference Becker, G.S. 1975 Human Capital, Second Edition. New York: Columbia University Press. Bojars, George. 1982. The Earnings of Male Hispanic Immigrants in the United States. Industrial and Labor Relations Review, 35(3): 343-353. April Borjas, George. 1989. Economic Theory and International Migration. International Migration Review. Vol. 23, No. 3. Pp: 457-484 Borjas, George. 1994. The Economics of Immigration. Journal of Economics Literature, Volume 32, and Issue 4 (Dec., 1994), Pp: 1667-1717 Chiswick, B.R. 1978. The Effect of Americanization on the Earnings of Foreign-Born Men. Journal of Political Economy, 86(5): 897-921. Oct. England, Paula, George Farkas, Barbara Kilbourne and Thomas Dou.1988. Explaining Occupational Sex Segregation and Wages: Findings from a Model with Fixed Effects. American Sociological Review, Vol. 53, No. 4. (Aug., 1988), pp. 544-558. England, Paula, Lori L. Reid, and Barbara Kilbourne. 1996. The Effect of the Sex Composition of Jobs on Starting Wages in an Organization: Finding from the NLSY. Demography 33: 511-521 Cobb-Clark, Deborah A. 1993. Immigrant Selectivity and Wages: The Evidence for Women. The American Economic Review, Vol. 83, No. 4. (Sep., 1993), pp. 986-993. 24

Goyette, Kimberly and Yu Xie (1999). The Intersection of Immigration and Gender: Labor Force Outcomes of Immigrant Women Scientists Social Science Quarterly 80: 395-408 Guo, Guang and John Hipp. 2003. Longitudinal Analysis for Continuous Outcomes: Random Effects Models and Latent Trajectory Models. Edited by Melissa Hardy & Alan Bryman, Sage Publisher North, David S. 1995. Soothing the Establishment: The Impact of Foreign-Born Scientists and Engineers on America. New York: University Press of America. Littell, Ramon, George A. Milliken Walter W. Stroup and Russell D. Wolfinger. 1996 SAS System for Mixed Models SAS Publications Long, Scott. 2001. from Scarcity to Visibility: Gender Differences in the Careers of Doctoral Scientists and Engineers. National Research Council Pedraza, Silvia. 1991. Women and Migration: The Social Consequences of Gender. Annual Review of Sociology 17: 303-325 Raudenbush and Bryk, 2002. Hierarchical Linear Models. Second Edition. Sage Publications Tang, Joyce. 1993. The Career Attainment of Caucasian and Asian Engineers. Sociological Quarterly 34(3): 467-496 Tang, Joyce. 2000. Doing Engineering: The Career Attainment and Mobility of Caucasian, Black, and Asian-American Engineers. Lanham, Md.: Rowman and Littlefield Publishers. Vernez, Georges. 1996. How Immigrants Fare in U.S. Education. Rand Report Xie, Yu, and Kimberlee A. Shauman. 2003. Women in Science: Career Processes and Outcomes. Harvard University Press. Zeng, Zhen and Yu Xie. 2004. Asian-Americans Earnings Disadvantage Reexamined: The Role of Place of Education. American Journal of Sociology. Vol 109. Pp: 1075-1108 25

Table1: Mean or Percentage of Variables by Survey Year, SESTAT Variable 1993 1995 1997 1999 Annual Salary 52636.58 55900.15 62250.95 68358.33 Annual Salary(logged) 10.749 10.661 10.860 10.894 Female 0.262 0.276 0.279 0.277 Asain 0.122 0.121 0.113 0.112 Other under represented minority 0.117 0.112 0.111 0.114 Foreign-born without citizenship 0.072 0.064 0.045 0.042 Foreign-born with citizenship 0.122 0.128 0.133 0.133 Native-born 0.805 0.808 0.821 0.824 Birth Year (1965 or later) 12.230 8.020 8.620 7.150 Birth Year (1955-1964) 32.760 31.740 31.480 31.850 Birth Year (1945-1954) 31.290 33.760 34.750 36.410 Birth Year (1935-1944) 17.390 19.270 19.250 19.650 Birth Year (1934 or earlier) 6.330 7.210 5.900 4.940 Full-time job 0.999 0.924 0.914 0.908 Bachelor 0.409 0.343 0.344 0.317 Master 0.215 0.213 0.222 0.219 Doctorate 0.333 0.393 0.372 0.397 Professional 0.043 0.045 0.049 0.050 Other Degree 0.000 0.007 0.013 0.017 Computer & math Science 0.095 0.089 0.089 0.089 Life & related Science 0.178 0.193 0.188 0.167 Physical & related Science 0.115 0.118 0.117 0.126 Social & related Science 0.181 0.195 0.194 0.211 Engineering 0.272 0.245 0.238 0.229 Non-S&E Degree 0.160 0.160 0.175 0.178 Academia 0.262 0.285 0.273 0.274 Government 0.150 0.131 0.131 0.125 Industry 0.588 0.585 0.597 0.601 Having children under 12 0.338 0.373 0.352 0.338 Number of Individuals 97689 71975 63626 47757

Table2: Mean or Percentage of Variables by Survey Year and Gender, SESTAT Variable 1993 1995 1997 1999 Male Female Male Female Male Female Male Female Annual Salary 56008.81 43146.49 60692.62 43338.1 67396.87 48934.36 74062.99 53495.47 Annual Salary(logged) 10.818 10.556 10.787 10.333 10.969 10.577 11.017 10.573 Asain 0.121 0.126 0.122 0.119 0.113 0.114 0.111 0.115 Other under represented minority 0.097 0.172 0.094 0.157 0.095 0.155 0.097 0.160 Foreign-born without citizenship 0.075 0.065 0.068 0.055 0.047 0.041 0.044 0.037 Foreign-born with citizenship 0.122 0.121 0.129 0.124 0.135 0.130 0.134 0.132 Native-born 0.802 0.814 0.803 0.821 0.818 0.829 0.822 0.830 Birth Year (1965 or later) 10.810 16.240 6.760 11.340 7.400 11.770 6.090 9.900 Birth Year (1955-1964) 31.800 35.460 30.370 35.320 30.070 35.140 30.230 36.060 Birth Year (1945-1954) 31.410 30.970 33.550 34.330 34.540 35.310 36.340 36.610 Birth Year (1935-1944) 18.710 13.670 20.960 14.840 21.110 14.420 21.650 14.430 Birth Year (1934 or earlier) 7.280 3.660 8.360 4.170 6.880 3.360 5.680 3.000 Full-time job 0.999 0.997 0.952 0.851 0.943 0.838 0.942 0.821 Bachelor 0.405 0.420 0.336 0.360 0.341 0.352 0.314 0.326 Master 0.205 0.245 0.201 0.246 0.209 0.256 0.206 0.253 Doctorate 0.345 0.299 0.412 0.343 0.391 0.323 0.419 0.342 Professional 0.045 0.035 0.046 0.039 0.049 0.046 0.050 0.050 Other Degree 0.000 0.000 0.005 0.012 0.009 0.023 0.012 0.029 Computer & math Science 0.091 0.108 0.086 0.096 0.086 0.095 0.087 0.094 Life & related Science 0.160 0.227 0.178 0.232 0.174 0.225 0.152 0.205 Physical & related Science 0.128 0.078 0.135 0.074 0.134 0.072 0.145 0.076 Social & related Science 0.136 0.307 0.145 0.327 0.144 0.322 0.163 0.336 Engineering 0.331 0.105 0.305 0.090 0.298 0.083 0.286 0.079 Non-S&E Degree 0.154 0.177 0.151 0.182 0.164 0.203 0.166 0.210 Academia 0.235 0.339 0.256 0.361 0.242 0.353 0.244 0.351 Government 0.149 0.154 0.130 0.131 0.131 0.130 0.124 0.128 Industry 0.616 0.507 0.614 0.508 0.628 0.517 0.631 0.521 Having children under 12 0.621 0.406 0.598 0.491 0.624 0.538 0.582 0.519 Number of Individuals 72077 25612 52099 19876 45892 17734 34511 13246

Table3: Correlation Matrix for Earnings Across Different Survey Years Salary93 Salary95 Salary97 Salary99 Salary93 1.000 Salary95 0.755 1.000 Salary97 0.713 0.799 1.000 Salary99 0.654 0.736 0.825 1.000

Table4: Coefficients and Standard Errors of Logged Earnings Measured at a Two-year Intervals by Gender from the SESTAT Model1 Model2 Model3 Variance Structure Male Female All Coeff. Std. Err Coeff. Std. Err Coeff. Std. Err Intercept 9.39 ** 0.014 8.993 ** 0.027 9.333 ** 0.012 t 0.08 ** 0.002 0.074 ** 0.003 0.079 ** 0.002 Female -0.159 ** 0.004 White(omitted) Asain -0.076 ** 0.008 0.008 0.015-0.053 ** 0.007 Other under represented minority -0.097 ** 0.007-0.038 ** 0.011-0.077 0.006 Native-born(omitted) Foreign-born without citizenship -0.084 ** 0.009-0.13 ** 0.019-0.1 ** 0.009 Foreign-born with citizenship 0.009 0.007 0.004 0.013 0.005 0.006 Birth Year (1934 or earlier)(omitted) Birth Year (1965 or later) -0.29 ** 0.011-0.087 ** 0.025-0.241 ** 0.01 Birth Year (1955-1964) -0.039 ** 0.009 0.114 ** 0.023-0.008 0.008 Birth Year (1945-1954) 0.1054 ** 0.009 0.196 ** 0.022 0.118 ** 0.008 Birth Year (1935-1944) 0.1641 ** 0.009 0.198 ** 0.024 0.165 ** 0.009 Full-time 0.232 ** 0.01 1.209 ** 0.013 1.229 ** 0.008 Bachelor(omitted) Master 0.11 ** 0.006 0.178 ** 0.011 0.127 ** 0.005 Doctorate 0.394 ** 0.006 0.526 ** 0.011 0.429 ** 0.005 Professional 0.549 ** 0.012 0.585 ** 0.023 0.562 ** 0.011 Other Degree 0.084 * 0.027 0.181 ** 0.034 0.116 ** 0.021 Social & related Science(omitted) Computer & math Science 0.179 ** 0.008 0.245 ** 0.015 0.201 ** 0.008 Life & related Science 0.033 ** 0.007 0.016 0.011 0.032 ** 0.006 Physical & related Science 0.085 ** 0.008 0.098 ** 0.016 0.093 ** 0.007 Engineering 0.22 ** 0.007 0.332 ** 0.016 0.243 ** 0.006 Non-S&E Degree 0.128 ** 0.008 0.138 ** 0.014 0.137 ** 0.007 Industry(Omitted) Academia -0.23 ** 0.005-0.218 ** 0.009-0.23 ** 0.005 Government -0.093 ** 0.006-0.013 ** 0.012-0.071 ** 0.005 Having 2 more children under age 12 Having no children under age 12-0.092 ** 0.005-0.044 ** 0.012-0.081 ** 0.005 Having one child under age 12-0.041 ** 0.006 0.005 0.013 0.03 ** 0.006 ρar(1) 0.2405 0.003 0.205 0.005 0.231 0.002 σ2e 0.6361 0.002 0.989 0.005 0.734 0.002-2LL 480975.3 214395.2 701720 AIC 480979.3 214399.2 701724 BIC 480997.8 214399.2 701743 Number of Subjects 78280 30736 109016 Number of Observations 204579 76468 281047 **P<0.001 *P<0.05